Abstract: In the massive machine type communication, user devices are allowed to randomly access the network and transmit small packets occasionally by grant-free transmission. Correspondingly, receivers are required to perform the blind multi-user detection without scheduling and pilots. The Bayesian blind detection algorithm based on message passing can solve the above problem, but the parallel iterative calculation consumes massive computing resources with high computational complexity and unstable convergence. An algorithm combining serial interference cancellation with Bayesian message passing is proposed to improve the performance of the blind multi-user detection. By iteratively reconstructing and canceling the interference of correctly recovered users, the signal to interference plus noise ratio at the receiver is improved, which enhances the error performance and reduces the computational complexity. Meanwhile, the convergence stability is promoted by damping and re-initialization mechanisms. Simulation results show that the proposed algorithm has obvious advantages over the parallel Bayesian blind detection algorithm in the blind multiuser detection.

Key words: Bayesian inference, multi-user detection, message passing algorithm, serial interference cancellation